mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Davide Libenzi <davidel@xmailserver.org>
To: Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: [patch] epoll use a single inode ...
Date: Mon, 5 Mar 2007 13:25:50 -0800 (PST)	[thread overview]
Message-ID: <Pine.LNX.4.64.0703051311380.4907@alien.or.mcafeemobile.com> (raw)


Epoll does not keep any private data attached to its inode, so there'd be 
no need to allocate one inode per fd. For epoll, the inode is just a 
placeholder for the file operations and could be shared by all instances.
I'd like to use the same optimization even for the upcoming file-based 
objects, so if you see problems let me know.
One that Al was pointing out was that an fstat(2) over an epoll fd would 
show the same st_ino. IMO that should be fine since an fstat(2) over an 
epoll fd is not something you want to do in any case and expecting 
meaningfull results.



Signed-off-by: Davide Libenzi <davidel@xmailserver.org>


- Davide



eventpoll.c |   36 ++++++++++++++++++++++++++++++++----
1 file changed, 32 insertions(+), 4 deletions(-)



Index: linux-2.6.20.ep2/fs/eventpoll.c
===================================================================
--- linux-2.6.20.ep2.orig/fs/eventpoll.c	2007-03-04 14:40:01.000000000 -0800
+++ linux-2.6.20.ep2/fs/eventpoll.c	2007-03-05 13:03:52.000000000 -0800
@@ -258,6 +258,7 @@
 		   int maxevents, long timeout);
 static int eventpollfs_delete_dentry(struct dentry *dentry);
 static struct inode *ep_eventpoll_inode(void);
+static struct inode *ep_create_inode(void);
 static int eventpollfs_get_sb(struct file_system_type *fs_type,
 			      int flags, const char *dev_name,
 			      void *data, struct vfsmount *mnt);
@@ -279,6 +280,9 @@
 /* Virtual fs used to allocate inodes for eventpoll files */
 static struct vfsmount *eventpoll_mnt __read_mostly;
 
+/* Placeholder inode for eventpoll fds */
+static struct inode *eventpoll_inode;
+
 /* File callbacks that implement the eventpoll file behaviour */
 static const struct file_operations eventpoll_fops = {
 	.release	= ep_eventpoll_close,
@@ -763,15 +767,18 @@
 	 * using the inode number.
 	 */
 	error = -ENOMEM;
-	sprintf(name, "[%lu]", inode->i_ino);
+	sprintf(name, "[%p]", ep);
 	this.name = name;
 	this.len = strlen(name);
-	this.hash = inode->i_ino;
+	this.hash = 0;
 	dentry = d_alloc(eventpoll_mnt->mnt_sb->s_root, &this);
 	if (!dentry)
 		goto eexit_4;
 	dentry->d_op = &eventpollfs_dentry_operations;
-	d_add(dentry, inode);
+	/* Do not publish this dentry inside the global dentry hash table */
+	dentry->d_flags &= ~DCACHE_UNHASHED;
+	d_instantiate(dentry, inode);
+
 	file->f_path.mnt = mntget(eventpoll_mnt);
 	file->f_path.dentry = dentry;
 	file->f_mapping = inode->i_mapping;
@@ -1555,6 +1562,11 @@
 
 static int eventpollfs_delete_dentry(struct dentry *dentry)
 {
+	/*
+	 * We faked vfs to believe the dentry was hashed when we created it.
+	 * Now we restore the flag so that dput() will work correctly.
+	 */
+	dentry->d_flags |= DCACHE_UNHASHED;
 
 	return 1;
 }
@@ -1562,6 +1574,17 @@
 
 static struct inode *ep_eventpoll_inode(void)
 {
+
+	return igrab(eventpoll_inode);
+}
+
+/*
+ * A single inode exist for all eventpoll files. On the contrary of pipes,
+ * eventpoll inodes has no per-instance data associated, so we can avoid
+ * the allocation of multiple of them.
+ */
+static struct inode *ep_create_inode(void)
+{
 	int error = -ENOMEM;
 	struct inode *inode = new_inode(eventpoll_mnt->mnt_sb);
 
@@ -1626,10 +1649,14 @@
 
 	/* Mount the above commented virtual file system */
 	eventpoll_mnt = kern_mount(&eventpoll_fs_type);
-	error = PTR_ERR(eventpoll_mnt);
 	if (IS_ERR(eventpoll_mnt))
 		goto epanic;
 
+	/* Create the single instance of inode for all eventpoll fds */
+	eventpoll_inode = ep_create_inode();
+	if (IS_ERR(eventpoll_inode))
+		goto epanic;
+
 	DNPRINTK(3, (KERN_INFO "[%p] eventpoll: successfully initialized.\n",
 			current));
 	return 0;
@@ -1642,6 +1669,7 @@
 static void __exit eventpoll_exit(void)
 {
 	/* Undo all operations done inside eventpoll_init() */
+	iput(eventpoll_inode);
 	unregister_filesystem(&eventpoll_fs_type);
 	mntput(eventpoll_mnt);
 	kmem_cache_destroy(pwq_cache);

             reply	other threads:[~2007-03-05 21:25 UTC|newest]

Thread overview: 40+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2007-03-05 21:25 Davide Libenzi [this message]
2007-03-05 21:52 ` Eric Dumazet
     [not found] <45ED046A.5010508@argo.co.il>
2007-03-06  7:27 ` Davide Libenzi
     [not found] <45ED1A3C.6030707@argo.co.il>
2007-03-07  0:37 ` Davide Libenzi
2007-03-07  0:51   ` Linus Torvalds
2007-03-07  1:01     ` Davide Libenzi
2007-03-07  1:27       ` Linus Torvalds
2007-03-07  1:47         ` Davide Libenzi
2007-03-07  6:52     ` Eric Dumazet
2007-03-07  7:15       ` Davide Libenzi
2007-03-07  7:16       ` Eric Dumazet
2007-03-07  8:56         ` Christoph Hellwig
2007-03-07 17:21         ` Linus Torvalds
2007-03-07 17:02       ` Linus Torvalds
2007-03-07 17:31         ` Eric Dumazet
2007-03-07 17:36           ` Eric Dumazet
2007-03-07 17:45           ` Linus Torvalds
2007-03-07 18:06             ` Eric Dumazet
2007-03-07 18:30               ` Linus Torvalds
2007-03-07 18:52                 ` Eric Dumazet
2007-03-07 19:07                   ` Linus Torvalds
2007-03-07 22:14                   ` Anton Blanchard
2007-03-07 22:57                     ` Linus Torvalds
2007-03-08  1:25                       ` Michael K. Edwards
2007-03-08  2:48                         ` Kyle Moffett
2007-03-08  7:24                           ` Eric Dumazet
2007-03-08 16:57                             ` Valdis.Kletnieks
2007-03-09  1:34                             ` Kyle Moffett
2007-03-09  2:52                               ` Anton Blanchard
2007-03-09  2:51                             ` Anton Blanchard
2007-03-08  3:20                         ` Linus Torvalds
2007-03-08  8:37                           ` Michael K. Edwards
2007-03-09  2:46                       ` Anton Blanchard
2007-03-08  8:56           ` Christoph Hellwig
2007-03-08  9:42             ` Eric Dumazet
2007-03-08 10:21               ` Christoph Hellwig
2007-03-08 11:11                 ` Eric Dumazet
2007-03-08 11:18                   ` Christoph Hellwig
2007-03-08 15:52                   ` Linus Torvalds
2007-03-08 20:00                   ` Bob Copeland

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=Pine.LNX.4.64.0703051311380.4907@alien.or.mcafeemobile.com \
    --to=davidel@xmailserver.org \
    --cc=akpm@linux-foundation.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=torvalds@linux-foundation.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®